[
https://issues.apache.org/jira/browse/HDFS-6879?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14111998#comment-14111998
]
Masatake Iwasaki commented on HDFS-6879:
----------------------------------------
Thanks for the review comments [~cmccabe]!
bq. SpanReceiverHost: can we call this "SpanReceiverManager"? It doesn't seem
to have anything to do with hosts. Similar for the config key.
I think "host" means not computer host but organizer here. Though I think
SpanReceiverManager is clearer, I would like to keep this unchanged in order to
use the same terminology already used in HBase.
bq. A higher-level question. There are also situations where we'd like to be
able to trace just a specific RPC. For example, we might find that writing to a
particular HDFS file is slow, and we'd like to be able to run some tests on
writing to this particular file while forcing tracing to be on. What's the plan
for this? Perhaps we could add a config option to the DFSClient to force
tracing to on, that could be used on "sample runs."
That is a good point. Implementing custom sampler and hooking it will make it.
I would like to add utility for the ease of tracing to DFSClient (and FsShell?)
in follow-up JIRAs.
Sampling in server-side might be a bit difficult. It needs server-side config
or sending filtering info via RPC.
> Adding tracing to Hadoop RPC
> ----------------------------
>
> Key: HDFS-6879
> URL: https://issues.apache.org/jira/browse/HDFS-6879
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Reporter: Masatake Iwasaki
> Attachments: HDFS-6879-0.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.2#6252)
